← All selected workReviewed 12 Nov 2026
Completed · Hackathon prototypeBehavioral banking loyalty prototype · Adria Hack Sarajevo 2026
G-Core
A prototype that rewards financial consistency measured against a customer's own history instead of spending volume. It scores five weighted habits with plain rules, protects unavoidable health costs, and ignores transfers that only move money between a person's own accounts.
- Organisation
- Team EMPTY · Adria Hack Sarajevo 2026 (24-hour hackathon)
- Role
- Team member and repository owner
- Timeline / Repo
- Sep 2026
Traceroute path · 5 hops
Each hop isolates problem, constraints, architecture decisions, verification artifacts, and limits.
- G-Core asks whether consistency, measured against a customer's own last three months, can work as a loyalty signal.
- The repository states plainly that this is a hackathon prototype and not an empirical claim about existing loyalty models.
Next hop follows below- Scoring is deterministic rather than machine learning, because there was no labelled data to train or validate a model responsibly.
- Raw transactions stay on the bank side; only a pseudonymous ID, points, tier and eligibility flags reach the loyalty layer.
Next hop follows below- Weights: budget discipline 30%, payment consistency 25%, savings consistency 20%, liquidity resilience 15%, goal consistency 10%. The repository labels these as design parameters, not calibrated coefficients.
- A protected-expense rule keeps a health cost out of discretionary scoring, so an emergency does not cost the customer their streak.
- An anti-gaming rule finds a transfer out and a matching transfer back between the customer's own accounts and excludes both from savings progress.
Next hop follows below- The screenshots below come from the demo, which uses a made-up customer and a made-up bank.
- The repository lists a GitHub Pages demo; I could not open it from my checking environment, so confirm it loads before sharing the link.

Monthly plan and tier ladder, computed from the customer's own last three months (demo data). 
A protected healthcare expense is excluded, so the streak is unaffected. 
An internal transfer between the customer's own accounts is detected and excluded. Next hop follows below- Not a credit score and not validated across customer groups; the weights are untested design choices.
- A production version would need open-banking connectors, authentication, key management, a privacy impact assessment and a fairness review.